SecurityAI App BuilderSupabaseReact Native

Is Your AI-Built App Secure? A 15-Point Security Checklist Before Launch

A 15-point security checklist for apps built with AI app builders and coding agents: database rules, secrets, payments, auth tokens, OTA updates and more.

DeveloperMill5 min read

AI app builders and coding agents make it easy to ship features. They don't make it easy to ship safe features. The code works on the happy path, and the gaps only show up when someone goes looking. In 2025, a single misconfiguration pattern left more than 170 apps from one AI builder with databases anyone could read (CVE-2025-48757). In 2026, one showcased app exposed 18,697 user records.

This checklist is what we run on every AI-built app before launch. It's written for founders. You don't need to fix every item yourself, but you should be able to answer each one.

Database and backend

1. Access rules are on for every table

If your app talks to Supabase, Firebase or a similar backend directly, the public key in your app is, by design, public. Database access rules are the only thing standing between that key and your data. In Supabase that means Row-Level Security (RLS) enabled on every table in an exposed schema. In Firebase it means security rules that aren't allow read, write: if true. Our Supabase RLS guide walks through the fixes.

2. Policies are tested as the wrong user

"RLS is enabled" isn't enough. Test every table three ways: logged out, logged in as the owner, and logged in as a different user. The third test catches the most common bug: a policy that checks "is logged in" instead of "owns this row".

3. Storage buckets have rules too

Uploaded files such as avatars, documents and receipts usually live in a separate storage service with its own permissions. Public buckets and guessable file paths are a common leak. Private files should be served through signed URLs or storage policies tied to the user.

4. Admin and service keys never leave the server

Supabase's service_role or secret key, Firebase Admin credentials and any key that bypasses access rules must only exist on your server or edge functions. If one is in your app code, assume it's already leaked: rotate it.

Secrets and the app bundle

5. No third-party secrets in the app

Search your project for API keys: AI providers, payment processors, email services, maps, analytics write keys. Anything bundled into an iOS or Android app can be extracted with free tools. AI-generated apps often call OpenAI or Stripe straight from the client because that's the shortest path to a working demo. Move those calls behind your own endpoint, which checks the user and then calls the provider.

6. Environment variables are understood

In Expo, any variable prefixed EXPO_PUBLIC_ is compiled into the app and readable by anyone. That's correct for a publishable key and wrong for a secret. Check every one.

7. The repo history is clean

If a secret was ever committed, deleting it in a later commit doesn't remove it from history. Rotate any key that was ever in the repo, especially if the repo was public or shared with an AI tool's cloud workspace.

Authentication

8. Tokens live in secure storage

Session and refresh tokens belong in the iOS Keychain and Android Keystore, which Expo exposes as expo-secure-store (our guide). AsyncStorage isn't encrypted. If you're using a Supabase or Firebase client, check which storage adapter it was configured with.

9. Sign-in flows are complete

Password reset, email verification, session expiry and sign-out on all devices are the edges AI tools skip. Deep links used for magic links and OAuth should be validated, so a malicious link can't push your app into a logged-in state. See adding authentication to an Expo app.

10. Users can delete their account

Apple requires apps that support account creation to let users delete their account from inside the app, and Google Play requires a deletion path too. Beyond compliance, deletion has to actually remove or anonymize the data, including files in storage.

Payments and business logic

11. The server decides who paid

Never unlock paid features based on something the app reports. Verify purchases through store receipts, RevenueCat or Stripe webhooks, and store entitlements server-side. Our in-app purchases guide covers the pattern.

12. Inputs are validated on the server

Client-side validation is for user experience. Anything that matters, such as prices, quantities, roles, limits and other users' IDs, must be checked by the server or database. A common AI-generated bug lets users edit their own role or plan column because the update policy allows any column change.

13. Expensive endpoints are rate-limited

AI features, SMS verification and email sending cost money per call. Without rate limits a script can run up your bill overnight. Add per-user and per-IP limits on anything that calls a paid API.

Release and operations

14. Updates are trusted

If you ship over-the-air updates with EAS Update, restrict who can publish to production channels and consider EAS Update's end-to-end code signing, so the app only runs bundles you signed. Keep dependencies current and run npm audit as part of CI.

15. You'll know when something goes wrong

Add crash and error reporting, and make sure it doesn't log personal data like tokens, passwords, full request bodies or health data. Set up alerts for spikes in auth failures and database errors. Every US state has a data breach notification law, so you need to be able to tell what was accessed and when.

How AI-built apps usually score

On a typical vibe-coded app, items 1, 2, 5 and 11 are the ones that fail most often, and they're also the most serious. The good news is they're fixable without a rebuild: policies can be rewritten, keys rotated and moved server-side, and payment checks added behind the existing UI.

If you're not sure where your app stands, that's what a production-readiness review is for. And if you're weighing the bigger question of whether to keep building with AI, read our 3-year cost comparison of hiring a developer vs an AI app builder.

Frequently asked questions

Are apps built with AI app builders secure?

Not by default. AI tools are good at producing working features and weak at access control, secret handling and server-side validation. CVE-2025-48757 showed more than 170 apps from one builder with databases readable without logging in. Treat AI-generated code as a draft and check it against a security checklist before real users sign up.

What is the most common security mistake in AI-built apps?

Missing or overly permissive database access rules. With backends like Supabase or Firebase, the app talks to the database directly using a public key, so Row-Level Security policies or security rules are the only protection. AI tools often leave them off, or write a policy that allows everyone.

Is it safe to put an OpenAI or Anthropic API key in my mobile app?

No. Anything shipped in an iOS or Android app bundle can be extracted. Put AI provider keys, payment secrets and admin keys on a server or edge function, authenticate the user there, and have the app call your endpoint instead of the provider directly.

Where should a React Native app store auth tokens?

Use the platform's secure storage: the iOS Keychain and Android Keystore. In Expo that's expo-secure-store. Don't store tokens in AsyncStorage or plain files, which aren't encrypted.

Bring us your AI prototype

Get a free 30-minute production-readiness review. We'll look at security, data access, store compliance and what it takes to ship, and you'll get a straight answer on whether to rebuild, harden or launch as-is.

Book your free review